Which curves are dangerous? A network-wide analysis of traffic crash and infrastructure data

Abstract

We conducted spatial analyses of traffic crashes, which took place in Czechia over 2010–2016, with respect to the road geometry data. The aim of the work was to identify hazardous road sub-segments where higher than expected numbers of traffic crashes occur.The entire Czech road network (58,200 km) was segmented at intersections into 39,074 between-intersection segments of varying lengths. Each road segment was further automatically sectioned, according to its horizontal alignment, into geometry-homogenous units (horizontal curves and tangents). Overall, 257,101 curves, defined as curved sections with radii below 2100 m, and 136,388 tangents, were identified. Subsequently, traffic crashes were joined to the respective geometrical units to determine their hazardousness. The degree of hazardousness was determined relatively, on a segment-by-segment basis, in order to eliminate the lack of precise traffic exposure data. In addition, the exact binomial test and Bayesian inference were used to identify the most hazardous horizontal curves.It was found that, in general, the curves with a higher crash risk have lower radii than the other curves. We identified the geographical locations of all curves with a high crash hazard. We also ranked the curves according to the crash hazard. Approximately ten percent of road segments contained at least one hazardous horizontal curve. 6943 significantly hazardous curves were identified by the use of the exact binomial test. The Bayesian inference reduced this number to 1395 (0.31% of the entire road network) and ranked them according to the Bayes factor. The most hazardous curve was 45 m long and contained 8.7 traffic crashes per year. Its hazard rate accounted for 37.4. This state-wide analysis of primary data was conducted over an extremely short time (up to 3 days) as theresult of an application of an efficient algorithm for automatic road curvature determination.